Adobe’s new scaling technology uses AI to sharpen video

Most of the new features and experiments that Adobe has announced recently involve AI, such as adding and removing objects for Premiere Pro and generating text images in Photoshop. Now the company has been revealed VideoGigaGANan experimental AI function that it says can increase video by eight times without the usual artifacts such as flickering or distortion, On the edge reported.

VideoGigaGAN beats other video super-resolution (VSR) methods because it avoids the usual artifacts and jitter introduced by GANs (General Adversarial Networks), according to Adobe. At the same time, it adds sharpness and detail – where most other systems fail to do both.

Of course, the system invents details that don’t exist from whole tissue, so it wouldn’t be suitable for things like forensic video enhancement, à la CSI style crime shows. But the details it adds look impressively real, like skin texture, fine hair, details of swan feathers and more.

According to the Adobe researchers, the model is based on a large-scale image upscaling program called GigaGAN. Previous VSR models had difficulty generating rich detail in results, so Adobe married “temporal attention” (reducing artifacts that build up over time), feature spreading (adding detail where nond exists), anti-aliasing and something called an HF shuttle to create the final result.

If added to products like Premiere Pro or After Effects, it could allow video producers to make low-resolution stills look much better, although using artificial intelligence to enhance humans is a controversial practice. No word yet on whether Adobe plans to do this (currently, clips are short and play at 12 frames per second), but many companies, including NVIDIA, Microsoft, Blackmagic Design and others, are also working on AI scaling technology.